A non-convex clustering shared memory folding compression method applied to digital office data clusters

With the widespread adoption of digital office systems, the rapid growth of data scale poses greater challenges to storage resources and access efficiency. Although traditional folding compression methods help reduce storage overhead, their decompression processes often introduce additional access latency, which in turn restricts overall system performance. Therefore, this paper proposes a shared memory folding compression method for digital office system data clusters, designed based on a non-convex clustering algorithm. A double-slope normal logarithmic model is employed to characterize the energy distribution characteristics of data clusters. An adaptive normalized least mean square filter is introduced into the high-efficiency cluster to construct a redundant data reduction model with low computational overhead and high prediction accuracy. Furthermore, data are divided into two parts through cost calculation, and a non-convex clustering model based on Gaussian kernel density estimation is established. The ability to identify data differences is enhanced by feature weighting, while the weight matrix of the deep neural network is approximately compressed using matrix singular value decomposition. In this paper, a performance gain analysis model for memory folding compression is established, and key metrics such as compression time, decompression time, and access frequency are comprehensively evaluated. Experimental results show that the proposed method can gradually increase memory utilization from 61.1 to 70.9%, demonstrating a significant compression effect. In the scenario of writing data in LZO format to a hard disk, processing 1.02 × 10⁷ pieces of data takes only 97 s, and data transmission speed is increased by approximately 31.6%.